The Agent Skills Directory

[Remote Code Execution] (MEDIUM): The PowerShell dependency check script recommends installing system tools by piping scripts from the internet directly into the interpreter. This pattern is often used in malicious context for initial access.
Evidence: Invoke-RestMethod -Uri https://get.scoop.sh | Invoke-Expression in examples/check-ffmpeg.ps1.
Evidence: iex ((New-Object System.Net.WebClient).DownloadString('https://community.chocolatey.org/install.ps1')) in examples/check-ffmpeg.ps1.
[Command Execution] (LOW): The core functionality relies on executing external binaries (yt-dlp, ffmpeg, jq) with arguments derived from user-provided URLs.
[Indirect Prompt Injection] (LOW): The skill processes untrusted metadata from external websites (such as video titles and descriptions) using yt-dlp --dump-json. This content could contain instructions targeting the AI agent.
Ingestion points: examples/download-scripts.sh (function show_info).
Boundary markers: Absent.
Capability inventory: Subprocess execution of shell commands.
Sanitization: No explicit sanitization of metadata output before processing.
[Privilege Escalation] (LOW): The helper scripts recommend the use of sudo for package installations on Linux/macOS, which is a standard requirement for system-wide dependency management.

yt-dlp